March 17, 2016

Market Structure Insights

Andrew Upward
Head of Market Structure
aupward@weedenco.com
203.861.7618

IEX: An Intriguing Wrinkle in the National Market System

EXECUTIVE SUMMARY
IEXs speed bump is a clever innovation that protects hidden liquidity providers from trading at stale
prices. The benefits to the liquidity provider spill over to the liquidity-taking side of the trade, which is
where we think institutional investors will find the most value. We applaud this effort to enhance
market structure for the benefit of institutional investors. Displayed liquidity on IEX is a different story.
Its tougher to figure out what types of traders will make use of it, and in what scenarios. Granting tradethrough protection to these orders presents costs that must be weighed against the benefits of allowing
IEX to compete as a full-fledged exchange. Valid counter-arguments notwithstanding, we feel that the
costs outweigh the benefits, and that the rules suggesting that IEXs quotes should be denied tradethrough protection should be upheld.
INTRODUCTION
No topic stirs up as much passion and curiosity among our clients as IEX, the Alternative Trading System
applying to become the 13th US stock exchange. Amid all the talk about the need for a holistic review of
the markets rules and structure, the folks at IEX went out and built a business that strikes at the heart
of many of the practices that frustrate large swaths of the buy- and sell-sides of the industry. In just two
and a half years, thanks to the good will theyve generated among this buy- and sell-side cohort and
thanks to the publicity boost from Michael Lewis best-selling book Flash Boys its grown to become
the second-largest ATS in the country.
Now that IEX is attempting to bring its model into the exchange realm, some of Wall Streets biggest
players have strenuously objected in a series of public comment letters. They argue that as an exchange,
IEX would violate some of the rules and conventions that underpin the National Market System the
web of exchanges, ATSs and broker-dealers that together facilitate all trading of US listed securities. IEX
vehemently disagrees, and adds that if it does violate any rules, theyre rules that other exchanges
violate too.
In this report, we first consider how the IEX model would work if the Securities and Exchange
Commission approves its exchange application as is, making allowance for the impact of its recently
amended routing scheme. We then discuss the rules and conventions that IEX would violate as a fullfledged exchange, and explore an idea for a compromise that we think has real merit: approving IEXs
application but denying its quotes trade-through protection. In the appendix at the end of the report,
we illustrate the IEX basics in a series of diagrams: what is the speed bump, how does it work, and
how does it add value to IEXs hidden peg orders and router.

Important disclosures appear on the final page of this document

March 17, 2016

PART I: THE EXCHANGE

In this section, we speculate about the types of liquidity that IEXs three main components the dark
pool, the displayed pool and the router are likely to attract or repel when it becomes an exchange. We
assume throughout the section that the SEC is going to approve the venues exchange application in its
current form. IEX is proposing to operate similarly to how it operates today as an ATS, so we dont
believe there will be any major surprises.
The router will look different, however. On February 29, IEX announced that it would be making an
adjustment to its order-handling procedures for routable orders. This adjustment, which we explain
below and diagram in the appendix, will likely reduce the flow of big marketable orders into IEX. And
that may have knock-on effects that constrain the venues potential market share.
Still, the pool of hidden peg liquidity has intrinsic value and should continue to attract a healthy balance
of liquidity providers and takers. The pool of displayed liquidity is a wild card. Our best guess is that it
will garner more interest from brokers than it does today, but that the HFT community will be slow to
embrace it.
THE EXISTING ROUTING SCHEME MAKES IEX BIGGER
Any router is mechanically capable of performing the tasks that IEXs router performs today. What sets
IEXs router apart is its ability to operate in the microseconds following an execution on IEX without the
chaos cancels, modifications, trades, re-pricings, etc. that often accompanies an execution. We
illustrate this chaos in Figures 9, 10 and 11 in the appendix, and in Figures 12, 13 and 14 we show how
the speed bump keeps the chaos at bay by creating a temporal buffer between the IEX execution and
subsequent executions at away markets.
No other exchange router can operate with such a buffer not in relation to executions taking place on
IEX certainly, and not in relation to executions on its own book either. No other ATS-operated router
that we know of operates this way. Broker-dealer-operated routers that dont work in tandem with an
ATS can schedule individual routes to arrive at away markets at around the same time, but they cant
offer such routes on the heels of a potential fill in an ATS.
This means that IEX alone offers what is essentially a package deal for aggressive liquidity consumers: a
free option to consume liquidity on the IEX book before going out to undisturbed away markets with any
unexecuted portion of the order. To the extent that this unique package succeeds in attracting
aggressive order flow, the IEX liquidity pool gets a valuable look at what are effectively parent orders
orders that have not yet been broken up into smaller child orders for routing to away markets.
For an agency broker looking for block-sized fills, providing liquidity in a pool known to attract bigger,
parent-sized marketable orders is a no-brainer. For an HFT market maker, intercepting a big order
before it gets broken up and sprayed to away markets could be more attractive than being on the
receiving end of the spray at those away markets especially if the intercepting is done with a speed
bump-protected hidden order (more on this below). So not only does the presence of bigger marketable
orders logically imply bigger fills and thus a bigger market share for IEX, the knowledge that those bigger

March 17, 2016

orders are coming through incentivizes providers to post in bigger size. As the saying goes, liquidity
begets liquidity.
THE NEW ROUTING SCHEME WILL LIMIT MARKET SHARE
As rival exchanges and some trading firms have pointed out, the unique value that the speed bump adds
to IEXs routing capabilities would seem to qualify as unfair discrimination between customers, issuers,
brokers, or dealers. Specifically, IEX would seem to be giving its own router, which will be registered as
a broker-dealer once IEX becomes an exchange, an unfair advantage over other broker-dealers. The
language cited above is taken from Section 6(b)(5) of the Securities Exchange Act of 1934, and any
exchange whose rules are found to permit such discrimination shall not be registered as a national
securities exchange.
Well never know whether the SEC would have interpreted IEXs routing advantage as unfair
discrimination, because IEX recently announced that it is changing how it handles routable orders. From
what we can glean from publicly available documentation, once a routable order has traversed the
speed bump, it will be sent directly to the router and not to the matching engine, which is where it is
sent today. From there, the router will look at the market data it is receiving from all thirteen
exchanges and decide on a schedule for the child orders it will send to IEX and away markets. The
market data it receives from IEX itself is the top-of-book feed, TOPS, and it is delayed by 350 s by a
speed bump that sits between the router and the matching engine. That speed bump also delays by 350
s any orders sent to the IEX matching engine by the IEX router. In Figures 15-18, we illustrate how the
new system architecture will work in the same urgent liquidity demand scenario used in Figures 9-14.
As IEX states in its most-recent comment letter, the new configuration will protect the sender of the
routable order from quote fading, just as the current configuration does. And the router itself will be
just as effective as it is today. The value that the router adds to the IEX liquidity pool, however, will
diminish. Orders resting on the IEX book will pair off against a child order, not the bigger parent order.
And the child order will only be as big as IEXs total displayed interest, plus any shares that IEX tacks on
in search of hidden liquidity when there arent enough displayed shares across all markets to satisfy the
full size of the marketable order.
We can only guess to what extent the new configuration will limit IEXs market share. Stats from IEXs
website show that for every 100 shares IEX matches on its own book, it routes about 30 residual shares
to away markets. These numbers dont tell us how much of the matched volume came from routable
orders, unfortunately. It could be that senders of routable orders are getting nothing done on IEX before
their orders are routed out, and the matched volume is all from non-routable orders. Or it could be that
the senders of routable orders are getting massive amounts done on IEX, and in fact all of its matched
volume is from these orders.
Obviously, the truth lies somewhere between these two extremes. We also need to bear in mind that
some brokers will continue to use the router for its ability to schedule child orders for maximum
displayed liquidity capture. Our gut tells us that the market share limitation will be significant, but not
severe. It certainly wont kill the venue.

March 17, 2016

HIDDEN PEGS SEGMENT THE MARKET
IEX the exchange is planning to operate its dark pool in much the same way that IEX the ATS
operates it today: as a pool of hidden peg orders that it re-prices using the other exchanges proprietary
data feeds. As we illustrate in Figures 6-8, once a hidden peg order traverses the speed bump and posts
to the book, it cant be accessed if an NBBO change has already happened or is imminent. This feature is
friendly to the liquidity provider, who is always going to be buying at a stable bid price or selling at a
stable offer price. Conversely, it discriminates against the liquidity taker who wants to trade but cannot
do so at least, not in those fleeting moments just before or after a quote change.
This is segmentation. It starts with the buy-side buying into the IEX value proposition and ethos and
requesting that the sell-side send orders IEXs way. That, along with the attractiveness of the router,
primes the pump with a steady stream of marketable and midpoint-pegged order flow. Those flows
encourage brokers to try to provide liquidity passively and to also treat IEX as a midpoint crossing
engine. But HFT market makers are also part of this story. Theyre drawn to any venue that offers a
chance to interact with marketable flow early in its lifespan, and the speed bump sweetens the deal by
ensuring that they only trade when theres a good chance of making money.
Furthermore, users of IEXs discretionary peg order a hidden order that pegs to the near side of the
NBBO but has discretion to execute at prices at or between the near side and the midpoint will be
restricted to executing only at the near side of the NBBO when IEX determines that the market is at risk
of moving against the user. This feature, which IEX calls Crumbling Quote protection, empowers traders
to provide liquidity at more aggressive prices by protecting them from trading at those prices when
doing so is likely, in IEXs view, to produce a losing trade from a short-term P&L perspective.
In addition, the fact that the discretionary peg order posts to the near side with discretion to execute up
to the midpoint as opposed to simply posting at the midpoint means that the user receives the price
improvement when the incoming order is priced to cross the spread. These two features Crumbling
Quote protection and discretion to the midpoint make the discretionary peg order type very attractive
to liquidity providers.
While we do believe that agency brokers and their buy-side clients derive value from the pool on the
provide side, they must share that value with a hungry HFT market-making community. The advantage
that the speed bump confers upon the hidden peg order is decidedly short-term: it protects against the
trade looking bad hundreds of microseconds from now. The advantage it confers is also predicated on
the trader being indifferent to whether or not she trades: theres no guarantee that, having avoided a
trade at a bad or stale price, shell be able to get the trade off at the new, better price. Institutional
investors holding periods are measured in weeks, months or years, and they cant always be indifferent
to whether or not they trade. So even though the hidden peg order has been heavily marketed to
institutions, and is useful to them, it is arguably better suited to HFT market makers. As such, we suspect
that those market makers will come to dominate the provide side if they dont already.
Wed be remiss, meanwhile, if we didnt highlight the advantages that accrue to the liquidity taker. We
stated earlier that the speed bump discriminates against the liquidity taker, and thats true but only
for a trader that attempts to take liquidity just before or after the NBBO moves in their favor. Traders
seeking liquidity at any other time will not only have an opportunity to cross against natural midpoint

March 17, 2016

and passive liquidity providers, theyll also find HFT market makers emboldened by the speed bumps
protection to provide more and/or better-priced liquidity than they might otherwise feel comfortable
providing on a non-speed bump market. As most retail brokers and wholesalers can attest, this is the
beauty of segmentation: when its done well, value accrues to both sides of the trade.
Were confident that the value the IEX dark pool is bringing to the marketplace through segmentation
will persist after it becomes an exchange. That said, we think institutional investors and their agency
brokers wont always be able to derive value from it as passive, non-block liquidity providers, and that
the most common interaction will be an agency broker taking liquidity from an HFT market maker.
We wonder too if the buy-side will continue to use the venue as a block crossing engine, and if so, for
how long. The block-crossing space is competitive, and theres nothing inherent in IEXs model that we
can see that gives it a concrete advantage over other pools. That will be especially true once IEX changes
its order-handling procedures for routable orders, and the flow of bigger marketable orders that
currently pass through the book on their way to the router is cut off. Of course, block-crossing is a game
of perception: as long as traders perceive it to be a venue for blocks, theyll keep sending their block
orders there and block trades will materialize.
THE DISPLAYED POOL: NOT HFT-FRIENDLY
Like a hidden peg order, a displayed limit order must traverse the speed bump on its way into the IEX
matching engine. Unlike a hidden peg order, once it arrives and is posted to the book, any subsequent
modification or cancelation of that order must be originated by the trader and must also traverse the
speed bump. In other words, IEX doesnt re-price the order on the traders behalf based on its lowlatency view of away-market quotes, as it does for its hidden peg orders.1
The result will likely be a displayed-liquidity pool that looks and feels much like other displayed pools to
agency brokers, but that looks and feels agonizingly slow to HFT firms (in this section, we use agency
brokers as shorthand for firms that dont co-locate their servers in multiple data centers, and HFT
firms as shorthand for firms that do).
First, consider the agency brokers experience. If its servers were in Times Square in Midtown
Manhattan, its latency experience would look something like this:

Note that IEX does re-price displayed orders that would lock or cross an away market upon entry, and
subsequently re-prices those orders so that they remain one tick less than the NBO (for buys) or one tick more
than the NBB (for sells). But this form of re-pricing is fundamentally different than IEXs re-pricing of hidden peg
orders. Also note that IEX offers non-pegged hidden limit orders that can only be modified or canceled by the
trader, and not by the exchange just like IEXs displayed orders.

March 17, 2016

To arrive at these distance-driven latencies, we simply converted the as the crow flies distance
between each pair of cities into time at the rate implied by IEXs speed bump (i.e. 350 microseconds per
38 miles). Depending on what kind of technology a firm is using, actual latencies could be higher or
lower than the ones weve come up with here.
As the diagram shows, the 368 s our agency broker has to wait before its order arrives at the IEX
matching engine, and the 368 s it has to wait before receiving market data from IEX, is not too
dissimilar from the latencies to and from the other big exchanges. If our broker is relying solely on feeds
from the Securities Information Processors (SIPs), it would receive IEX quote and trade updates
around the same time that it would receive updates from Bats via the SIPs (this isnt shown in the
diagram). All in all, speed bump notwithstanding, the experience of sending and modifying and canceling
displayed orders on IEX will feel very similar to the brokers experience at other exchanges.
Now lets consider an HFT firms perspective. If the firm is co-located in all three major data centers
Secaucus, Mahwah and Carteret it will have near-zero latency to all the major exchanges. There may
be scenarios or strategies for which consolidating data from the three locations is necessary, in which
case the Mahwah-to-Carteret latency (313 s using our speed bump-based conversion factor) would
serve as the firms minimum latency. Put differently, from the perspective of the firms Mahwah servers,
NYSE data will be real-time but Nasdaq and Bats data will always be slightly stale. From the perspective
of its Secaucus servers, Bats data will be real-time but NYSE and Nasdaq data will always be stale, and
from the perspective of its Carteret servers, Nasdaq data will be real-time but NYSE and Bats data will be
stale. And theres no way for the firm to have the best of all worlds, no way to patch together and see
all three feeds in real-time at the same time.
Regardless of whether the firms strategy relies on consolidated data, the latency from IEXs speed
bump could be problematic. Lets say that the HFT firm is offering stock at the same price on IEX, Nasdaq
and Bats, and its offer on Nasdaq gets lifted. Its next move is to try to raise the price of its offers on IEX
and Bats to avoid unwanted exposure. It routes a cancel/replace message to Bats in Secaucus from its

March 17, 2016

servers in Carteret, and within about 150 s the message has arrived and Bats has re-priced the order.
At the same time, it routes a cancel/replace message to IEXs point-of-presence in Secaucus, and within
about 150 s it enters IEXs system architecture and begins to traverse the speed bump en route to the
matching engine.
Any order message that arrives at the front door of IEXs point-of-presence in Secaucus after the HFT
firms cancel/replace message has gone through it would trail the HFT firms message as both race
around and around the coiled fiber that is the speed bump. So as long as the HFT firm gets its message
to the front door before any marketable buy orders, the HFT firm will successfully raise the price of its
offer before it can be traded against. In this way, the HFT firms displayed-order experience on IEX would
be similar to its displayed-order experience on every other exchange.
The only problem for the HFT firm is that for a period of time that could be as long as 850 s in our
example the 150 s it took to get the cancel/replace message from Carteret to the IEX front door in
Secaucus, plus the 350 s it would take for the cancel/replace message to travel to the IEX matching
engine, plus the 350 s it would take for IEX to send an acknowledgement back it doesnt know if it
has succeeded or failed in getting its offer out of the way. That 850 s compares unfavorably to the 150
s it took to find out the status of its offer on Bats.
Any time an HFT firm has a limit order in the marketplace and doesnt know its status, that firm is
uncertain of its exposure and thus cannot take action to reduce it. Granted, were talking about
extremely small windows of uncertainty here hundreds of microseconds and of course brokers that
dont co-locate their servers live with these windows of uncertainty today on every order. Indeed, even
the fastest, most-sophisticated HFT firms that are co-located in every major data center currently
endure some uncertainty when they need to wait for information from one venue before acting on
another. And thats not going to change unless the fragmented market model itself changes.
These caveats aside, we suspect that many HFT firms wont want to tolerate the added uncertainty that
comes from not knowing the status of their IEX displayed limit orders at crucial moments. We also note
that IEXs proposed rule book seems to rule out Day Intermarket Sweep Orders and makes no mention
of a post-only modifier for displayed limit orders. These are powerful order types that all of the
incumbent exchanges offer, and that HFT firms use to exert tighter control over their orders. The Day
ISO instructs the exchange to execute and/or post the order regardless of its view of away markets, and
the post-only modifier instructs the exchange to reject or re-price the order if it would take liquidity
before posting to the book. We suspect that HFT firms would find the post-only modifier particularly
useful on IEX, as the odds that a non-marketable order becomes marketable en route to the exchange
are higher given the extra latency created by the speed bump.
It also appears as though IEX intends to retain TOPS as its only proprietary data feed. TOPS only shows
aggregated top-of-book order information, which would make it difficult for HFT firms to know where
their orders are positioned in IEXs displayed order queue. Knowing where their orders are positioned in
the queue helps these firms manage their risk.
For all these reasons, we dont expect to see much HFT liquidity in IEXs displayed pool, particularly not
when these firms can provide liquidity more safely by using the hidden peg order. We think theyll
continue to make displayed markets and pursue arbitrage opportunities on the incumbent exchanges,

March 17, 2016

while ignoring IEX as a displayed market and treating its dark liquidity as a niche trading opportunity
much like they do today.
THE DISPLAYED POOL: AGENCY BROKERS MAY FIND POSTING ATTRACTIVE
Agency brokers could be well served by posting displayed child orders on IEX when working the parent
orders passively. Theyre accustomed to latency, so the speed bump wont necessarily scare them off.
And then there are a host of potential benefits:
1. Their orders would have priority over equally-priced hidden orders.
2. If displayed liquidity from HFT firms is scant, as we expect, their orders would rest higher up in
the displayed order queue than on a typical exchange. All else equal, being higher up in the
queue increases the odds of capturing spread.
3. It sounds as though IEX is planning on discounting the trading fees for posting and consuming
displayed liquidity (they hint at this in their first and second comment letters).
4. The rumored low take fee might attract some incremental marketable flow.
5. Only displayed orders will be capable of attracting child orders from the IEX router under the
new order-handling procedures for routable orders.
Against these benefits, agency brokers would have to weigh the risk of executing at soon-to-be-stale
prices. But really, that risk exists on all the exchanges, and its probably more acute on high-rebate/hightake fee exchanges. Meanwhile, as we hinted earlier in the report, trading at stale prices is a shortterm alpha concept that isnt always relevant for an agency broker that actually needs to trade so that
its investment-manager client can build or liquidate a position. Posting displayed on IEX is a surprisingly
intriguing idea for agency brokers, though the value of such a strategy will diminish to the extent the
new routing scheme curtails the stream of marketable flow into the pool.
Clearly there are a lot of moving parts to the proposed IEX exchange. Predicting just how they might fit
together is beyond our powers of foresight, although we hope our observations and speculations have
provided some food for thought. We wont really know what type of liquidity ecosystem IEX is capable
of generating until it goes live as an exchange.

PART II: THE IMPACT

Until now, weve focused solely on IEX: who benefits from it today, and who might benefit from it in the
future. In this section, we consider the broader impact of IEX exchangehood on market quality and the
overall US equity trading experience.
First, some perspective: porting IEXs less-than-2% market share from the ATS realm into the exchange
realm is not going to change much. Even if it gains market share rapidly, well still be looking at an NMS
that is roughly two-thirds exchanges, one-sixth ATSs and one-sixth everything else give or take a few
percentage points. It will continue to hum as a fast, fragmented, algo-dominated system.
What were looking for are fissures in the dam: small weaknesses that could grow bigger over time. To
that end, we see two issues with IEX gaining full-fledged exchange status that could potentially be
problematic:

March 17, 2016

1. Granting IEXs quotes trade-through protection could prevent the NMS from becoming more
efficient in the future
2. IEXs model could set a precedent of exchanges interfering with brokers pursuit of best
execution
While we focus on these negatives in the remainder of the report, its only fair that we salute IEX for its
entrepreneurial spirit, its innovative model that delivers real value to the marketplace, and its
commitment to improving market structure.
TRADE-THROUGH PROTECTION: MORE COSTS THAN BENEFITS
Whether motivated by the threat to their own business models, or by a genuine desire to preserve the
integrity of the US stock market or both exchanges and trading firms have roundly criticized IEXs
attempt to bring its model into the exchange realm. The ensuing back-and-forth between IEXs
supporters and its detractors has shined a light on aspects of IEXs model that appear to violate existing
rules and conventions. The most controversial of these is whether IEXs displayed quotes and orders
should be afforded trade-through protection.
IEXs detractors claim that the Commission is opposed to giving trade-through protection to exchanges
that have deliberately introduced latency into their order-handling processes. As proof of their claim,
they point to a single sentence in the 523-page Reg NMS adopting release:
The term "immediate" precludes any coding of automated systems or other type of
intentional device that would delay the action taken with respect to a quotation.
If IEXs speed bump is deemed to be such an intentional device, it would not be acting immediately
to execute, cancel or acknowledge quotes and orders. If its reactions to quotes and orders are not
immediate, its own displayed quotes should be manual and not automated per Reg NMS Rules
600(b)(3) and 600(b)(37). Manual quotes are not afforded trade-through protection, which means that
brokers wouldnt have to connect to or trade on the IEX exchange (although a desire to achieve best
execution might oblige brokers to connect).
IEX has tried valiantly in its comment letters to argue that the speed bump is not actually an intentional
device, but that looks like wishful thinking to us. The problem is that the Commission makes it clear that
anything can be an intentional device that is in fact an intentional device and that delays action taken
with respect to a quotation and it seems to us that the speed bump meets these qualifications.
IEX also argues that if the speed bump is an intentional device, then so are the incumbent exchanges
points of presence, and so are the delay coils they use to equalize latency among their co-location
customers. All three of these features, IEX contends, amount to prescribed physical distance between
the customers servers and the matching engine. Thus, if all three features are essentially the same, and
one of them (the speed bump) is deemed an intentional device, then they should all be so deemed.
These points also seem wide of the mark. NYSE and Nasdaq and Bats dont force brokers to connect via
their points of presence as IEX does, which means those exchanges dont force brokers to experience
latency as IEX does. Any broker willing to pay for co-location services can ignore NYSEs or Nasdaqs
points of presence completely. Bats doesnt offer co-location services, which means that brokers can

March 17, 2016

park their servers next to the Bats matching engine in Secaucus without paying Bats anything beyond
the $400/month port fee that all customers must pay. The fact that Bats offers a point of presence in
Weehawken for the convenience of brokers whose servers are located there does not mean that it has
built latency into its system not in the way that IEX has at any rate. More to the point, nothing about
the distance between a point of presence and its corresponding matching engine looks remotely like an
intentional device.
Going back to Bats, its worth noting that although Bats itself doesnt offer co-location services, Equinix,
which owns the NY5 data center in which the Bats matching engine is located, does. And Equinix doesnt
offer it up for free. We understand that Bats negotiated to have Equinix offer co-location services to
Bats customers at a discounted rate, but this only underscores the value of the space immediately
surrounding exchange matching engines. Bats decided that it didnt want to be in the business of leasing
out that space; NYSE and Nasdaq decided that they did want to be in it. Either way, somebody owns and
is leasing out the space at rates that exceed the going rates elsewhere.
In other words, co-location costs more because the space is inherently worth more. IEX ingeniously uses
the latency of the speed bump to devalue the space around its point of presence (and its matching
engine), and this is what levels the playing field. This is how IEX achieves fairness. But thats only true for
its hidden orders. For its displayed orders, the space around its point of presence still has value because
getting order messages there with haste can improve outcomes (see the above section on IEXs
displayed pool not being friendly to HFT firms for more on this). In this regard, IEX is simply going the
Bats route and letting Equinix profit by leasing the space out.
Incidentally, if the SEC wanted to level the playing field without IEXs ingenious but latency-adding
innovation, it could do so by strictly regulating all connectivity charges. This would include lease rates for
server space in all locations. That kind of incursion into a highly developed corner of the private sector
would have ramifications, of course, and we suspect that the SEC doesnt want to go there. But its an
option.
As for the delay coils that the incumbent exchanges use to equalize latency among their co-location
customers, its true that theyre also intentional devices that delay action taken with respect to
quotations. The SEC would be justified in revoking the incumbents protected-quote status for the same
reason that it would be justified in denying that status to IEX. We dont think the SEC would do that,
however. The magnitude of the latency introduced by the delay coils is tiny single-digit microseconds
and therefore we highly doubt it impacts any firms trading experience or routing decisions.
Furthermore, the coils actually make it possible for the exchanges (and data centers) to offer co-location
services to all comers on equal terms. Without them, the brokers with servers closest to the matching
engine would have an advantage over the brokers with servers on the outer edges of the data center.
And the exchanges could conceivably charge them different rates. What is now a two-tier market would
explode into a ten- or twenty-tier market.
To summarize, we think IEXs speed bump is an intentional device, and we think it is uniquely an
intentional device. That said, it doesnt interfere with the accessibility of IEXs displayed orders. As we
explained above in the section on IEXs displayed pool not being HFT-friendly, traders racing their orders
to the IEX point of presence get access to the displayed liquidity that exists at the moment those orders

10

March 17, 2016

arrive even though it takes another 350 microseconds for the event (trade, cancel, modification etc.)
to actually materialize.
The speed bump only interferes with the timing of the event and the subsequent dissemination of
acknowledgment messages and market-data updates. IEX will disseminate market-data updates to the
SIPs without delay, but of course the SIPs incur a separate latency when consolidating and redistributing
the data in their feeds. The regulator may treat this direct-to-SIP dissemination as IEX fulfilling its
obligation to ensure promptness of reporting per Rule 601 of Reg NMS, but its of little practical
significance to the latency-sensitive trading community that glues the markets together most tightly.
If the SECs stated opposition to intentional devices stemmed only from a concern over the
accessibility of a displayed quote or order, we could see it ignoring the fact that the speed bump is an
intentional device and granting IEXs quotes trade-through protection. However, if any part of its
opposition stemmed from a concern over the Reg NMS adherence problems caused by stale market
data an increase in locked and crossed markets, pegged orders on other venues trading at stale prices,
gaming, etc. then the Commission would have grounds for denying that protection.
If the Commission did deny IEXs quotes protected status out of concern over Reg NMS adherence
problems, it would be an awkward admission that the SIP feeds are superfluous at least when it comes
to helping market participants and venues avoid Reg NMS adherence problems. (If the SIP feeds were
useful in this regard, then the fact that IEX is pledging to send its market data to the SIPs directly not
via the speed bump would be enough to assuage any concerns.) Most market participants already
know that the SIP feeds arent useful in this way its an open secret but it might be awkward for the
Commission to have to acknowledge the fact.
On the other hand, if the Commission grants IEXs quotes protected status, it will have to contend with
copycat rule filings from any incumbent exchanges wanting to play the speed bump game. Complexity
would be a real concern, and the Commission would need to contemplate guiding the exchanges to
acceptable types and magnitudes of latency. Otherwise, exchanges could attempt to introduce speed
bumps that cater to certain segments of the market in a way that harms market quality.
In its Reg NMS re-proposal in December 2004, the Commission was clear that it didnt think guiding
exchanges to acceptable magnitudes of latency was the right course of action (from section II(A)(2)(a)
here):
The definition of automated quotation does not set forth a specific time standard for
responding to an incoming order. The Commission agrees with commenters that the standard
should simply be "immediate" i.e., a trading center's systems should provide the fastest
response possible without any programmed delay.
Thats not to say that the SEC couldnt chart a new course, with the rationale being that the market has
changed considerably since those words were written. But 2004 wasnt that long ago. It will want to
consider whether any wisdom still lingers in its initial decision. We suspect that the primary concern,
then as now, is that capping the allowable latency at a certain level would create a latency floor below
which exchanges would lose competitiveness. For this to happen, the market share and revenues gained

11

March 17, 2016

by building a bigger pool of speed-bump-protected hidden liquidity would have to outstrip the share and
revenues lost by having a less-competitive displayed quote.
Were not convinced that all exchanges would go for such a trade-off, but to the extent they did, latency
would proliferate and price discovery would become less efficient. It goes back to HFT firms not coping
well with the uncertainty over the status of their outstanding orders. If more exchanges implemented
speed bumps, the uncertainty would become harder and harder to avoid, and eventually bid-ask
spreads could widen out. A lot of dominoes would have to fall for this to happen, but that is the risk.
The SEC could decide not to give explicit guidance around latency types and magnitudes, and consider
each copycat filing separately. But by accepting speed bumps it liked and rejecting those it didnt, it
would simply be providing implicit guidance instead.
With or without explicit guidance, a decision to give IEXs quotes protected status looks to us like a
recipe for more complexity and more Reg NMS adherence problems, and potentially more latency and
less-efficient price discovery. The irony of the situation is that no one knows if IEXs displayed orders will
add any value. The speed bump-protected hidden pool is the component that we know is innovative and
that levels the playing field between those who can and cant afford co-location. The displayed orders
will be uncompetitive almost by definition (because of the latency), so its not clear to us what benefits,
if any, would materialize if the SEC were to give them trade-through protection.
Meanwhile, being denied trade-through protection would come with a nice silver lining: rival exchanges
would be less likely to roll out speed bumps of their own, because doing so would mean sacrificing their
protected-quote status. That would leave IEX to dominate and perhaps expand the niche its already
carved out for itself.
COMPLICATING BEST EXECUTION
Brokers are in the business of accepting orders from clients and executing those orders in a way that
helps the client meet its trading objectives. This process may include crossing stock with another
institution, or executing in dark pools or on exchanges. Brokers use their knowledge of the execution
options available to them, and their understanding of the clients objectives, to decide where to send
the clients orders.
Exchanges, on the other hand, are public markets that exist mainly to bring together buyers and sellers.
Historically, they have done so without judging or anticipating the value of the trade to the buyer or the
seller. The no judgment aspect helps ensure that a market is always available to parties who arent
looked upon favorably by a particular broker, but who nonetheless have a legitimate trading interest.
Technology and regulation have evolved in such a way that broker-dealers can easily create quasiexchanges (ATSs), while exchanges can and do perform broker-dealer tasks like order-routing. But
regulators have always made sure that where the lines between brokers and exchanges blur, they dont
disappear entirely.
Our concern is that as IEX transitions from broker-dealer to exchange, it will bring with it broker-like
functionality that all exchanges will then be able to mimic. All brokers would then be disadvantaged
relative to exchanges, which could use their inherent advantages to compel brokers to outsource those
broker-like functions to them. IEXs old routing scheme was the clearest example of broker-like activity

12

March 17, 2016

that might have been irresistible when combined with other elements of the exchanges offering. That
advantage is no longer in play, but our concerns havent been completely assuaged.
For not-held orders, brokers have the freedom to route where they please, when they please, subject to
their duty to act in accordance with the principles of best execution. If they want to outsource certain
functions that theyre ultimately responsible for, like keeping a buy order pegged to the bid side of the
market, they can use an exchange order type to do so. But the responsibility of executing at a good or
fair price ultimately lies with the broker, not with the exchange. The same goes for an exchanges
router: using one does not absolve the broker of its own best execution obligation.
If a broker wants to locate its servers in Secaucus next to the Bats matching engines, because it feels
that that location provides the best optimization of cost, speed and centrality for its purposes, it can do
so. If instead it wants to locate them in farm country in the middle of Pennsylvania because the space
there is much cheaper and the latency doesnt impede its pursuit of best execution, it can do so. If it
wants to co-locate in all three of the major data centers because thats what it feels is needed to achieve
best execution, it can do so.
The mix of freedom and responsibility assumed by the broker works well because clients have different
trading objectives. Some objectives require latency-sensitive strategies, and some dont. Some are pricesensitive, and some are time-sensitive. Some require natural liquidity, and some can be achieved by
trading solely with intermediaries. Clients gravitate toward the brokers that serve them best.
Giving the broker wide latitude to pursue best execution may not always lead to the best outcomes, but
we feel that it works better than the proposed alternative: exchanges that have no knowledge of the
end-clients trading objectives manipulating the brokers message traffic, and preventing the broker
from pursuing executions on away markets, in order to produce outcomes that they think will be good
for the end-client. And of course the question looms large of how an exchange can protect one
brokers client without harming another brokers client when those two end-clients would otherwise
have traded with one another, but were prevented from doing so.
We want to be clear that this is not an attack on IEX itself. Our concern is that in exchange form, and as a
precedent-setter for other exchanges, it threatens to limit brokers freedom to pursue best execution as
only they can pursue it.
CONCLUSION
At its core, IEX is a trading venue that segments out late-arriving marketable order flow, and it does this
by holding up brokers message traffic. Its ingenious, it benefits those who trade on it, and we value it
as an ATS but it shouldnt be an exchange with a protected quote.
If it were approved as an exchange with an uprotected (manual) quote, it would take some getting
used to, but we think that it might actually be a solution that works for everyone. IEX would gain SelfRegulatory Organization status and could take an active role in NMS governance. Its clearing costs would
go to zero and it would gain immunity from liability when carrying out SRO functions. Most importantly,
its business model wouldnt have to change at all it would still protect investors orders, and those of
HFT market makers, in the same way that it does today.

13

March 17, 2016

The downside for IEX is that it would forgo any claim on the quoting share of the tape revenues, which
accrues to SROs with protected quotes only. And any hopes it may have had to attract non-marketable
limit orders from retail brokers, or corporate listings, would probably be dashed.
Back on the plus side, however, rival exchanges would likely be dissuaded from proposing speed bumps
of their own. That would mean less competition for IEX, but, more importantly, it would assuage
concerns about complexity and meaningful latency seeping into the NMS. Reg NMS adherence problems
related to IEXs stale market data would be less troublesome, and brokers would not be forced to
subjugate their own interpretation of best execution to IEXs (but could still route to IEX when their
interpretations were aligned).
Intriguingly, amid calls from many corners of the industry to do away with the order-protection rule, the
SEC would be able to gather data on how often and under what circumstances IEXs unprotected quotes
were accessed or ignored. While not perfectly reflective of a market with no order protection
whatsoever, the data might still provide some insight into the pros and cons of taking such an action.
Approval with manual quotes makes a lot of sense to us. However, we recognize that many in the asset
management community want to see IEX in full-throated competition with the incumbent exchanges
despite the negative effect that it may have on market quality. And we should be clear that any negative
effect would be marginal given that were talking about microsecond timescales. Its probably more
accurate to say that the latency introduced by IEXs speed bump could prevent HFT firms from wringing
the last remaining drops of efficiency out of our fragmented market structure. Put differently, order
messages traveling at the speed of light with no friction upon entry into the exchange matching engine
would be the maximum-efficiency state that may not be achievable in a market dominated by speed
bumps.
All else equal, maximum efficiency is desirable. But it requires profit-seeking activity on the part of HFT
firms to make it happen, and the suspicion is that the biggest HFT firms are extracting rents from the
system not profits. Some asset managers and brokers hope that an IEX exchange with protected
quotes can change this dynamic, and restore the institutional investors pride of place at least to some
degree, without unduly harming market quality. Were not convinced that the trade-off will play out
that cleanly, but we certainly understand the desire to achieve a better, fairer market structure.

14

March 17, 2016

APPENDIX: HOW THE SPEED BUMP WORKS

The IEX ATS is often described as a dark pool, and for the most part thats accurate: most of its
liquidity is comprised of hidden orders. Some of these hidden orders are block-sized midpoint peg
orders, making IEX effectively a block-crossing engine for institutions. But its average trade size of
around 215 shares, per data from FINRA, suggests that there are plenty of smaller sized algorithmic
orders in the pool as well.
IEX is also a lit pool, as it accepts displayed orders. Any broker-dealer consuming IEXs proprietary data
feed, TOPS, can see the aggregated size of its top-of-book orders in real-time. IEX began accepting
displayed orders in late February 2015, and recent stats from IEXs website suggest that these orders
account for 8-10% of all shares executed on the venue.
Finally, IEX allows brokers to use its smart-order-routing technology to send the residual pieces of orders
not filled at the venue itself to the exchanges with the best prices. We refer to this technology
throughout the report as the router or IEXs router.
In isolation, IEXs dark pool, lit pool and router function much like other venues and routers. What we
and other market participants find valuable and controversial is how IEX employs a speed bump to
delay message traffic into and out of these components.
COILED OPTICAL FIBER THAT DELAYS MESSAGE TRAFFIC
IEXs speed bump is a coil of optical fiber through which order messages (including cancel and modify
messages) must travel before reaching the IEX matching engine. Messages that IEX sends back to the
customer, including messages sent to customers through its proprietary data feed, travel through the
coil in the opposite direction. Its approximately 38 miles long when stretched out, but it fits into a
compartment no bigger than a shoebox when coiled. It takes a message approximately 350 s
(microseconds) to traverse the coil.
The speed bump is physically located in a data center in Secaucus, New Jersey, just inside the perimeter
of IEXs system architecture. The system architecture is comprised of all of the information technology
systems that the exchange employs to accept, acknowledge, execute and route orders, to send out and
receive market data, to do risk checks, and to manage all the other processes that a trading venue must
manage. Customers connect their own servers to IEXs servers in Secaucus, and order messages that
customers send in buy, sell, cancel, modify, etc. must traverse the speed bump on their way out of
Secaucus en route to the matching engine a few miles away in Weehawken, New Jersey.
IEXs system architecture is unique in the sense that customers arent allowed to connect via servers in
the same data center as the matching engine a practice known as co-location. They can only connect
via its servers in Secaucus, which along with the speed bump are known as IEXs point of presence or
POP. To illustrate the difference between a more traditional exchange architecture and IEXs speed
bump-enabled, co-location-disabled architecture, weve created simplified representations of both in
Figures 1 and 2.

15

March 17, 2016

Figure 1.

Figure 2.

These schematics are of course only representative, and are certainly not to scale. That said, we believe
that they convey an accurate understanding of the salient features of IEXs system architecture based on
the information IEX has provided in its marketing material and in its comment letters.
ADDING REAL VALUE TO IEXS HIDDEN PEG ORDERS
The speed bump is designed to prevent fast, opportunistic traders, whom well call high-frequency
traders or HFT firms for simplicity, from trading against IEX orders that are pegged to an outdated or
stale (or soon-to-be-stale) National Best Bid or Offer. Trading at stale prices represents a kind of
bait-and-switch for liquidity providers in the sense that the description of, for example, a primary peg
order type leads them to believe that theyll be capturing the entire bid-ask spread, when in fact they
may only capture part of it or in some cases none of it.
To understand how IEX protects hidden peg orders and bear in mind that all IEX peg orders are hidden
consider an HFT firm that sees that the National Best Bid (NBB) in a given stock has just ticked down.
If IEX had no speed bump, this firm might quickly route a sell order to IEX in the hopes of finding a
hidden primary peg buy order to sell to at the old NBB before IEX itself sees that theres a new NBB.
If such a buy order existed on IEX, and if the HFT firm got their sell order to IEX before the venue knew
to lower the price of its buy order, the trade would go off at the price of the old NBB. The HFT firm
would have established a short position (or closed out a long position) at a higher price, and IEXs hidden
peg buyer would have paid more for the stock than they bargained for. The following diagrams illustrate
how such a scenario, referred to as latency arbitrage, might unfold if indeed IEX did not have its speed
bump.

16

March 17, 2016

Figure 3.

Weve brought in NYSEs data center in Mahwah, New Jersey to create a hypothetical two-exchange
market, and weve excluded the IEX POP and router as theyre not relevant in this scenario. At time zero,
NYSEs 12-cent bid is the NBB and IEX has a hidden buy order pegged to the NBB.
Figure 4.

Suddenly, at time t, a cancelation or trade on NYSE drops the NBB from 12 cents to 11 cents.

17

March 17, 2016

Figure 5.

Almost instantaneously following the NBB change at time t, an HFT firm routes a 12-cent sell order to
IEX. Even though IEX is taking NYSEs proprietary data feed in this hypothetical example, the HFT firms
superior technology allows it to get its order to IEXs matching engine before the quote update message
arrives. IEX thinks that the NBB is still 12 cents, so it executes the sell order against its hidden peg buy
order at 12 cents. A fairer price from the buyers perspective might have been 11 cents, the true NBB at
the time of the execution.
Now lets take a look at how this scenario would have unfolded if IEXs speed bump was operational,
which of course is the case today.
Figure 6.

Notice that weve added points of presence in Secaucus for both IEX and NYSE, and that the HFT servers
that clustered around the IEX matching engine in our earlier hypothetical now cluster around the IEX
POP. At time zero, NYSEs 12-cent bid is the NBB and IEX has a hidden buy order pegged to the NBB.

18

March 17, 2016

Figure 7.

Suddenly, at time t, the NBB drops from 12 cents to 11 cents.

Figure 8.

By forcing the HFT firm co-located at NYSEs data center in Mahwah to route its sell order to IEX via its
speed bump in Secaucus, IEX ensures that the quote update coming directly from Mahwah arrives first.
As shown by the red S, the HFT firms sell order is still spinning around the coiled optical fiber when IEX
gets word from NYSE that the new NBB is 11 cents. In this scenario, because the HFT firm sent the order
with a 12-cent limit, and IEXs best bid is now 11 cents, no trade will take place on IEX.
Its worth noting that the speed bump also offers protection against slightly different strains of latency
arbitrage than the one weve diagrammed above. It prevents executions against hidden IEX orders
pegged to stale NBBO midpoint prices, and it prevents a trader who suspects that a price is about to
become stale from picking off a hidden order on IEX that is pegged to that price. The latter scenario is
probably better described as aggressive or opportunistic trading, but its still referred to by many
somewhat sloppily in our opinion as latency arbitrage.

19

March 17, 2016

ADDING REAL VALUE TO IEXS ROUTING CAPABILITIES
The IEX router is effectively co-located in its Weehawken, New Jersey data center as depicted by the
green server icon in Figure 2 above. Any residual piece of a marketable, routable order that isnt filled
against IEXs own orders is sent without delay to the router with instructions to route the shares out to
the exchanges with the best prices.
IEX doesnt publicly disclose the methodology by which it routes those residual shares, but given the
routers high fill rates and IEX executives well-documented knowledge of the latencies that exist
between the exchanges system architectures, we can safely assume that the basic idea is to schedule
the individual routes so that each arrives at its destination at around the same time as all the others.
Doing so would ensure that liquidity providers at one venue cant pull or change the price of their orders
in response to executions taking place at another venue a practice known as quote fading.
We dont need to know the exact methodology of the router to know that the speed bump makes it
more effective. While the above examples showed how the speed bump delays inbound message traffic,
it delays outbound message traffic as well. The outbound delay prevents all market participants and all
other venues who consume IEXs TOPS feed from knowing that an execution has just taken place at IEX.
Because the traders posting liquidity at other markets have no chance to pull or adjust their quotes in
response to news of the execution (i.e. no ability to fade), the IEX router has no need to race its orders
out to these markets. It can take its time, knowing that the liquidity that it sees now is in all likelihood
the liquidity that will be there a few hundred microseconds from now.
To illustrate how the speed bump empowers the router, we consider a scenario in which a trader has an
urgent need to purchase stock and leverages the IEX router to do so. Figures 9, 10 and 11 show what
would happen if IEX chose not to employ its speed bump, and Figures 12, 13 and 14 show what would
happen if IEX employed its speed bump as it does today.
Figure 9.

Note that weve added the IEX router to the schematic (see green server icon). At time zero, both IEX
and NYSE are offering stock at 15 cents.

20

March 17, 2016

Figure 10.

At time t, a routable 15-cent buy order for 5,000 shares arrives at the IEX matching engine and matches
against 3,000 shares on the IEX book, leaving a 2,000-share residual. Note that IEX was only showing
1,000 shares, but had 2,000 shares in hidden liquidity behind it.
Figure 11.

At time t, IEX immediately sends a trade message to the buyer whose servers are in Secaucus, and also
updates its market-data feed (m and d in the schematic). At the same time, it sees that NYSE is
offering 2,700 shares at 15 cents, so it instructs its router to route the 2,000-share residual there.
Although the instruction is given almost instantly at time t, the IEX router cant out-race the co-located
HFT firm that is reacting to the execution by routing a cancelation message to NYSE. It might be that the
HFT firm sold stock to the buyer in the 3,000-share transaction and doesnt want to risk selling more to
that buyer on NYSE, or it could be that the HFT firm is simply unnerved by the presence of what seems
like a large, aggressive buyer. In any event, by the time the IEX-routed buy order arrives at NYSE at time

21

March 17, 2016

t + 200 s, the amount of displayed interest has dropped to 1,700 shares not enough to fill the
balance of the buyers order.
Now we consider how the same scenario would unfold if IEX were operating with its speed bump in
place and without co-location. This is a representation of how IEX operates today.
Figure 12.

At time zero, both IEX and NYSE are offering stock at 15 cents. Note that the speed bump is now in place
and that the HFT servers are now clustered around the IEX POP, which is the only point of entry for
order messages into the IEX matching engine (and the only point of exit for proprietary market data and
trade messages sent from Weehawken).
Figure 13.

At time t, a routable 15-cent buy order for 5,000 shares arrives at IEX and matches against 3,000 shares
on the IEX book, leaving a 2,000-share residual.

22

March 17, 2016

Figure 14.

Still at time t, IEX immediately sends trade messages to the buyer and seller whose servers are in
Secaucus, and also updates its market-data feed. At the same time, it sees that NYSE is offering 2,700
shares at 15 cents, so it instructs its router to route the 2,000-share residual there. When the order is en
route to NYSE, no one but IEX itself knows that an execution has just taken place on IEX. The trade
messages and the proprietary data-feed updates are still traversing the speed bump en route to the
servers in the POP (m and d in the schematic), and the updates that IEX sends without delay to the
Trade Reporting Facility (not shown in the above diagram) wont be processed in time to alert the HFT
firms co-located in NYSEs Mahwah data center of the trade event. As a result, the buyer is able to buy
all 2,000 shares at NYSE at time t + 250 s, completing its 5,000-share parent order.
THE NEW ROUTING SCHEME WILL LIMIT MARKET SHARE
On February 29, IEX announced that it will change how it handles routable orders once it becomes an
exchange. Under the new scheme, once a routable order has traversed the speed bump, it will be sent
directly to the router not to the matching engine, which is where it is sent today. From there, the
router will look at the market data it is receiving from all thirteen exchanges and decide on a schedule
for the child orders it will send to IEX and away markets. In Figures 15-18 below, we illustrate how the
new system architecture will work in the same urgent liquidity demand scenario used in Figures 9-14.

23

March 17, 2016

Figure 15.

As before, we have a simplified two-exchange market with 1,000 displayed shares offered at 15 cents on
IEX and 2,700 displayed shares offered at the same price on NYSE. Note that there is now a second
speed bump that sits between the IEX matching engine and the IEX router.
Figure 16.

At time t, a 5,000-share buy order with a 15-cent limit arrives at the IEX router. The IEX router sees
the shares offered on IEX and at NYSE. Knowing that it will take about 350 s for a child order to arrive
at the IEX matching engine and about 250 s for a child order to arrive at the NYSE matching engine, it
gives the IEX child order a head start.

24

March 17, 2016

Figure 17.

At time t + 100 s, a child order is en route to the IEX matching engine via the speed bump (this is the
green B1), and the router releases the NYSE-bound child order (B2). Because the router only sees
3,700 displayed shares on NYSE and IEX, it has discretion over how to route the 1,300 shares that arent
spoken for. We assume that IEX will oversize both child orders 500 shares extra for the IEX child order
and 800 shares extra for the NYSE child order in the hopes of finding hidden liquidity.
Figure 18.

At time t + 350 s, both child orders arrive at their destinations. The buyer has purchased only 4,200
shares not enough to complete its 5,000-share order because hidden liquidity was found at IEX but
not at NYSE. At t + 700 s, the buyer will receive news of the execution on IEX, and 250 s later, at t +
950 s, the buyer will receive news of the execution on NYSE by way of IEX (these messages arent
shown in the diagram). Since it took 350 s to get the order to the router in the first place, users of IEXs

25

March 17, 2016

routable orders can expect to wait a total of 1,050 s for news of any executions on IEX. That compares
to the 700-s wait that users of routable orders must endure today, and that users of non-routable
orders must endure today and in the future. The wait for news of executions at away markets is a bit
longer with the new model too, since the router must hold up any child orders sent to those markets to
ensure that they arrive at around the same time as any child orders sent to the IEX matching engine
(assuming the router does send a child order to IEX).

Disclosure: This publication is prepared by Weeden & Co.'s trading department, and not its research department. This publication is
for information purposes only and is based on information and data from sources considered to be reliable, but it is not guaranteed
as to accuracy and does not purport to be complete and are subject to change without notice. This publication is neither intended
nor should be considered as an offer or the solicitation of an offer to sell or buy any security or other financial product. Nothing
contained herein is intended to be, nor shall it be construed as, investment advice. Information contained herein provides
insufficient information upon which to base an investment decision. Any comments or statements made herein do not necessarily
reflect those of Weeden & Co. LP or its affiliates. 2016 Weeden & Co. LP.

26